AI is great unless you let it think for you. It can help us scale thinking very effectively. Always ask it for sources for claims and check those sources. Always read what it has output thoroughly and manually edit before posting.
I hate AI slop but I do like it when it is used intelligently.
AI is like a popularity contest. It gives you the results of things with the most hype. If you look at that list, you can see it’s which LE supplements get the most press.
It’s not a list of what is the best. It’s a list of the most talked about on the internet.
In this case, AI is looking at a specific list of drugs and supplements. The problem is, that in evaluating the worth of those it bases its analysis on what the most common and hyped claims are. It doesn’t look for what claim is most valid, only which one is repeated most often. So, as an example, say, vitamin D had an obscure benefit of protecting your brain from aging (I’m making this up), but even though it’s very well documented, only one study has found this brain benefit. Meanwhile the greatest number of claims for vit. D is that it “helps bones”. Guess what AI is going to do - yep, it’ll vomit back to you claims about the bone, not because that’s best documented or most important, but because it’s the most popular and frequent. It’s a fatal flaw. Because not only does it force you to wade through a pile of commonly repeated garbage trying to validate it, but because it will likely not even include that one single study on brain protection, because it doesn’t know how to evaluate that claim… all the AI evaluations I’ve ever seen are the most simple minded following of some previously established procedure regardless of ultimate validity. We are still a very long way to it being useful in ways people already think of it as being. In other words, people give it more credit than it deserves. People assume AI abilities that may only come sometime in the future. We’ll see!
That’s a much better list. How did you compile it?
I built a prompt using this thread as inspiration. Different Ai gave me different lists.
I used Claude Ai and this prompt:
" Prompt:
You are a Longevity Research Analyst and Science Journalist.
Create a rank-ordered list of the Top 10 molecules/supplements to slow biological ageing, ranked purely on overall human outcomes evidence quality. Do NOT rank based on any specific mechanistic framework.
Ranking criteria (in order of weight):
- Hard human endpoints — all-cause mortality, cardiovascular death, hospitalisation, functional independence
- RCT quality and sample size — powered trials, not pilots
- Mammalian lifespan data — specifically ITP (NIA Interventions Testing Program, multi-site)
- Epidemiological signal strength — cohort size, confounder control
- Mechanistic breadth — secondary consideration only
For each molecule provide:
- 2–3 sentence rationale for its rank
- Mechanisms: 3–5 bullet points
- Evidence: table with 5 key studies (Title, Year, Species, 1-sentence finding)
- Safety: explicit statement — “Safe in short-term human trials”, “Known risk X at dose Y”, or “Data Absent”
Style: Objective, critical, no hype. Explicitly challenge weak evidence (e.g. “only rodent data”, “surrogate endpoints only”, “retrospective with healthy-user bias”, “conflicting trials”). Prefer human hard endpoints over mechanistic claims. If a molecule has strong rodent or mechanistic data but weak human outcomes data, say so clearly and rank it accordingly.
End with: a summary rankings table and an honest editorial conclusion about the state of evidence across supplements vs. approved drugs.
Gemini came up with this list, using the same prompt.
| Rank | Molecule | Human Endpoint | ITP Status | Quality of Evidence |
|---|---|---|---|---|
| 1 | SGLT2i | Mortality | Positive | High (RCTs) |
| 2 | Metformin | Mortality (Diabetic) | Neutral | High (Epidemiology) |
| 3 | Acarbose | CV Events | Positive | Moderate (RCTs) |
| 4 | Rapamycin | Biomarkers Only | Positive | Moderate (Pilots) |
| 5 | Omega-3 | CV Mortality | Neutral | High (RCTs) |
| 6 | Vitamin D3 | Cancer Mortality | Neutral | High (Meta-analysis) |
| 7 | Astaxanthin | Biomarkers Only | Positive | Low/Moderate (ITP-led) |
| 8 | GlyNAC | Functional Pilots | Mixed | Low (Sample Size) |
| 9 | Urolithin A | Muscle Function | N/A | Moderate (Function) |
| 10 | Lithium | Epidemiology | Neutral | Low (Observational) |
I think thats a pretty good list, as a starting point for people. It seems that exactly how you ask the question (your prompt) and the LLM system you ask, drives a wide variation in the end result you get.
I ran the same prompt on my Google Pro account and got a very similar response:
| Rank | Molecule / Intervention | Greatest Strength | Greatest Weakness |
|---|---|---|---|
| 1 | SGLT2 Inhibitors | Hard mortality RCTs + ITP lifespan | Side effect profile (Rx only) |
| 2 | GLP-1 Agonists | Massive cardiovascular mortality reduction | Lacks direct mammalian longevity data |
| 3 | Statins | Decades of human mortality reduction data | Targeted mechanism (lipids/CVD) |
| 4 | Metformin | Huge human epidemiological dataset | Conflicting human exercise/mammalian data |
| 5 | Omega-3 (EPA) | Strong secondary prevention mortality data | Mixed primary prevention results |
| 6 | Rapamycin | The undisputed gold standard in mice (ITP) | Zero human hard endpoint mortality RCTs |
| 7 | Acarbose | Excellent male mouse ITP data | Lacks primary prevention human longevity RCTs |
| 8 | Taurine | Unprecedented multi-species lifespan data | Zero human RCTs for longevity endpoints |
| 9 | GlyNAC | Reverses human aging biomarkers in RCTs | Zero human hard endpoint mortality RCTs |
| 10 | NAD+ Precursors | Mechanistic theory and marketing | Zero human outcome data; surrogate endpoints only |
A stark bifurcation exists between approved pharmaceuticals (ranks 1-4) and over-the-counter supplements (ranks 8-10). Because aging is not recognized as a disease by regulatory bodies, powered human clinical trials—which require hundreds of millions of dollars—are exclusively funded for pathology-specific drugs. Consequently, molecules like SGLT2 inhibitors and GLP-1 agonists possess indisputable human mortality data, whereas the supplement industry relies entirely on extrapolating from mice and marketing surrogate markers. If the objective is to base longevity interventions strictly on verifiable human outcomes rather than mechanistic promises, repurposing approved cardiometabolic drugs currently holds a vastly superior evidence base compared to any heavily marketed longevity supplement.
I sometimes think of AI as a hive mind/focus group summarizer. This list probably reflects the
balance of internet attention/noise rather than evidence. However it’s useful for that alone.
AI can be a great tool for signposting for attention though, which this thread is a lot about. Urolothin A for example - I’m interested to see replication by other labs. And the ai signposted me to a 2025 Whitefield study I was unaware of
Using the above suggested prompt (You are a Longevity Research Analyst and Science Journalist), the latest version of Claude came up with a somewhat different list.
And the latest GPT, also in deep thought mode. (Reproducing only the summary tables. Some of the discussion texture is more interesting.
I was surprised to see SGLT2 inhibitors at the top of the list. From my perspective, I take Jardiance (empagliflozin), the most popular drug in this class, and in my case, it is perhaps the poorest option for fasting glucose control that I have taken. Its glucose-lowering effect was barely noticeable. To achieve such a high ranking, it must be attributable to its other properties:
"While they were originally developed to lower blood sugar, large-scale clinical trials have proven they provide significant protection for the heart and kidneys, often even in patients who do not have diabetes.
SGLT2I are the most helpful to your kidneys of any of the most popular longevity medicines or supplements currently recommended.
These AI lists would be pretty good if they didn’t include metformin
Longevity and health wise, if I really just have to choose 10 and 10 only
I had GPT-5.5-thinking pool together everyone’s list above and reason about which from the lists would be its “top 10” pick, and it gave:
I’d probably draw up a different list, and would focus on combinations like “magnesium glycinate” and “potassium citrate”, that way you get several compounds in one.
Addendum: I asked it which ones it would add that aren’t on anyone’s list, and it came up with Cocoa flavanols / cocoa extract, Vitamin B12, iodine, potassium-enriched salt substitute, Resistant starch or inulin-type prebiotic fiber, a strain-specific probiotic.
I take all of those except ergothioneine. Aside from the longevity benefits, what are you expecting from that? Which product are you taking?
i just put my genome into Claude and asked it for my personal/bespoke list. Very interesting top 10 and very different to a standard top 20 list. Ive found the logic much more compelling for my bespoke, genomic-supporting list
Did you submit the raw datafile or a report?
Likely not an issue but the gene expressions that would suggest moderation in glycine consumption are not rare. Have you checked your profile? Also, a few of these supplements, especially at these doses, present a load on the kidneys. I assume you are young and have high kidney function.
